17:15Machine-translated from Azerbaijani by AIRead in Azerbaijani
1Password Launches Password-Hiding Login System for AI Agents
“1Password” has launched the “1Password for Agents” platform for artificial intelligence agents. The system allows agents to log in to accounts without seeing the username or password. After introducing its “Claude” integration in July, the company opened the system to other agents as well. This makes it possible to assign agents tasks that require account access, such as making restaurant reservations, buying gifts, and planning vacations. Before using an account, an agent must obtain the owner’s consent.
When an agent reaches a login page, it authenticates itself with a unique digital key. “1Password” then displays a notification showing which account credentials are being requested and why. After the user approves the request on their personal device, the app automatically fills in the login credentials without showing them to the agent. The process is similar to two-step verification. The agent continues its task without seeing the password. Permission applies only to that session and the information approved in advance. Consent must be obtained again in the next session. Passwords and one-time MFA codes—that is, additional security codes—are entered through a protected connection managed by the app. The agent can also perform consecutive operations across several websites. For example, when planning a business trip, the user will not be asked to reenter the password at every step. The system currently works with “Meta Muse,” “ChatGPT Work,” “Dots,” “Claude Duet,” “Grok Bot,” and “Town.” “Browserbase,” “Browser Use,” “Dana,” “Instinct,” “Kernel,” “Perplexity,” “Ramp,” “Tomo,” and “Wajo” are also preparing support for connecting to the system. According to the company, other services will be added to the list in the future. The user separately approves the access request for each agent. Usernames, passwords, and MFA codes are not shown to the artificial intelligence model. “1Password” says this information never leaves the user’s devices. Although the new system reduces the risk of a password being exposed to an agent, it does not completely prevent the agent from making mistakes. For example, an agent could order the wrong product or make an incorrect, nonrefundable travel booking. It is therefore important to check the purpose of the request and the selected account before granting permission. “1Password” has not yet announced the feature’s price or when it will become available to all users.
When an agent reaches a login page, it authenticates itself with a unique digital key. “1Password” then displays a notification showing which account credentials are being requested and why. After the user approves the request on their personal device, the app automatically fills in the login credentials without showing them to the agent. The process is similar to two-step verification. The agent continues its task without seeing the password. Permission applies only to that session and the information approved in advance. Consent must be obtained again in the next session. Passwords and one-time MFA codes—that is, additional security codes—are entered through a protected connection managed by the app. The agent can also perform consecutive operations across several websites. For example, when planning a business trip, the user will not be asked to reenter the password at every step. The system currently works with “Meta Muse,” “ChatGPT Work,” “Dots,” “Claude Duet,” “Grok Bot,” and “Town.” “Browserbase,” “Browser Use,” “Dana,” “Instinct,” “Kernel,” “Perplexity,” “Ramp,” “Tomo,” and “Wajo” are also preparing support for connecting to the system. According to the company, other services will be added to the list in the future. The user separately approves the access request for each agent. Usernames, passwords, and MFA codes are not shown to the artificial intelligence model. “1Password” says this information never leaves the user’s devices. Although the new system reduces the risk of a password being exposed to an agent, it does not completely prevent the agent from making mistakes. For example, an agent could order the wrong product or make an incorrect, nonrefundable travel booking. It is therefore important to check the purpose of the request and the selected account before granting permission. “1Password” has not yet announced the feature’s price or when it will become available to all users.